Research map / 3D vision / Face and shape
NormalGAN: Learning Detailed 3D Human from a Single RGB-D Image
Lizhen Wang, Xiaochen Zhao, Tao Yu, Songtao Wang, Yebin Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add a GAN conditioned on normal maps to the depth denoising network (ablation) | NormalGAN front-view depth denoising · Depth denoising, synthetic test set | MAE: 4.75 -> 4.63 | Improved |
| condition the GAN on normal maps rather than depth maps (ablation) | NormalGAN front-view depth denoising · Depth denoising, synthetic test set | MAE: 5.02 -> 4.63 | Improved |
| add a depth-conditioned GAN (DepthGAN) over no GAN (ablation) | NormalGAN front-view depth denoising · Depth denoising, synthetic test set | MAE: 4.75 -> 5.02 | Hurt |
| use NormalGAN instead of DDRNet retrained on the same data | depth denoising · Depth denoising, synthetic test set | MAE: 5.00 -> 4.63 | Improved |
| add simulated Kinect noise data augmentation (ablation) | NormalGAN · Real Kinect v2 captures (qualitative) | denoising and back-view stability | Improved |
| add normal-map-conditioned GAN for back-view depth inference (ablation) | NormalGAN back-view depth · Back-view depth inference (qualitative) | geometric detail | Improved |
| use orthographic re-projection instead of perspective view (ablation) | NormalGAN · Body boundaries and extremities (qualitative) | reconstruction quality around boundaries | Improved |
| add shading removal using spherical harmonics from refined depth (ablation) | NormalGAN front and back color · Real captures color inference (qualitative) | back-view color quality | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DDRNet — DDRNet: Depth Map Denoising and Refinement for Consumer Depth Cameras Using Cascaded CNNs (ECCV 2018) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ANIM: Accurate Neural Implicit Model for Human Reconstruction from a single RGB-D Image (CVPR 2024 · similarity 0.88)
- Normalized Avatar Synthesis Using StyleGAN and Perceptual Refinement (CVPR 2021 · similarity 0.88)
- Just Look at the Image: Viewpoint-Specific Surface Normal Prediction for Improved Multi-View Reconstruction (CVPR 2016 · similarity 0.87)
- Progressive Learning of 3D Reconstruction Network From 2D GAN Data (TPAMI 2023 · similarity 0.87)
- Dense 3D Object Reconstruction from a Single Depth View (TPAMI 2018 · similarity 0.87)
- Marr Revisited: 2D-3D Alignment via Surface Normal Prediction (CVPR 2016 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
